How Modeling Assumptions Reflect Values and Shape the World

For the same problem you can adopt different modeling approaches, each stemming from distinct underlying assumptions.

Assumptions reflect the modeler's values. In the past I only judged whether an assumption was clear and technically sound; now I realize that assumptions also embody value judgments.

Assumptions Determine How We See the World

Every model relies on assumptions that appear technical—data distribution, solvability, computational ease—but they actually reveal our understanding of how the world works.

For example, when studying educational equity, the first step is choosing an indicator:

Measuring fairness by "resource input" emphasizes balanced school hardware.

Measuring fairness by "outcome disparity" focuses on student achievement distribution.

Measuring fairness by "equal opportunity" examines family background and admission chances.

Each choice reflects a different notion of fairness and thus shapes model structure, variables, and objectives.

Choosing a Model Is Expressing a Stance

Technically we have many models—linear, hierarchical, Bayesian, optimization, machine learning—but selecting one is never neutral.

Using an optimization model implies a belief in "maximizing efficiency".

Using a game‑theoretic model assumes rational agents seeking maximal payoff.

Using a simulation model emphasizes multi‑agent interaction and complex evolution.

The same phenomenon modeled with different tools yields different answers and subtly reinforces distinct mindsets and value preferences.

What Is a "Good Assumption"?

Beyond clarity and verifiability, a good assumption should align with the values we expect the world to embody—fairness, respect, sustainability—avoiding structural discrimination.

Examples: an insurance pricing model that excludes high‑risk patients improves actuarial stability but marginalizes vulnerable groups; a recommendation algorithm that assumes users only like similar content creates an "information bubble" that harms public discourse.

Thus, a truly good assumption requires both technical soundness and value rationality.

Modeling Not Only Analyzes Reality, It Constructs Reality

Modeling describes the world, yet it also shapes it.

Examples:

City traffic optimization models decide which roads are widened and whether new bus routes are added.

Medical resource allocation models determine which patients receive priority care.

Education evaluation models influence how schools teach and how students learn.

When a model becomes part of a decision‑making system, its structure and outcomes become action guidelines, effectively directing the world according to the modeler's value choices.

Values Are Not Redundant, They Are Core to Modeling

All models originate from some value orientation; the difference lies in whether we acknowledge and honestly confront that orientation.

For instance, a traffic model may suggest canceling certain bus routes to improve average travel time, but if we prioritize equitable access for the elderly and low‑income residents, we might reject that recommendation.

Similarly, a medical allocation model that maximizes treatment success by favoring patients with higher recovery chances may conflict with the societal belief that every life has equal value.

A responsible modeler should not hide their values but make them explicit, even if it means sacrificing a bit of precision for greater social interpretability and ethical soundness.

Reflective questions to ask during modeling:

Do my data already contain bias?

Will my model amplify existing inequalities?

Who will benefit from the model’s output, and who might be harmed?

Beyond accuracy, what other concerns matter?

Not Only Math, Also Human Understanding

Recognizing modeling as a value practice turns modelers into value builders, requiring three non‑technical qualities:

1. Ethical sensitivity : anticipate unfairness or harm and embed ethical constraints.

2. Plural perspectives : respect diverse goals of different groups and avoid one‑size‑fits‑all designs.

3. Social responsibility : consider the externalities of model outcomes and aim to serve the public interest rather than merely efficiency or profit.

Technical skills can be taught, formulas can be looked up, algorithms can be transferred, but values must be cultivated, chosen, and should guide how we model the world.

Modeling Ultimately Is About Values

The models we build, the objectives we set, and the variables we deem important are expressions of our stance, belief, and values.

As the author Wang Haihua writes, "Modeling is fundamentally about values—how you model decides how you want the world to be understood and changed."